What is it about?
Anomaly detection based on generative models has become more important in recent years, especially in domains where abnormal data is scarce, but still suffers from significant problems. Existing reviews have focused on reviewing the state of the art but they have often neglected to try to summarize the underlying problems. This work focuses exactly on these challenges and describes which progress has been made in tackling the challenges while also describing open research areas in this field - and practical as well as theoretical problems researchers might discover.
Featured Image
Photo by Jukan Tateisi on Unsplash
Why is it important?
This work is - to the best of our knowledge - the first work describing the fundamental challenges of GAN based anomaly detection. Many findings are also applicable to similar generative models (e.g. diffusion models or autoencoders). The work helps researchers to orient themselves and avoid common pitfalls while also allowing investigations into the feasibility of utilizing GANs to tackle their problems.
Perspectives
The work helps researchers to understand fundamental limitations as well as opportunities of utilizing generative adversarial networks for their problems. It is - as far as I know - the first review which focuses on the discovery and description of these challenges.
Fiete Lüer
Gofore Oyj
Read the Original
This page is a summary of: Anomaly Detection using Generative Adversarial Networks
Reviewing methodological progress and challenges, ACM SIGKDD Explorations Newsletter, March 2024, ACM (Association for Computing Machinery),
DOI: 10.1145/3655103.3655109.
You can read the full text:
Contributors
The following have contributed to this page
